I tested a 4-rule and 12-rule mean-reversion strategy. The complex version looked better in-sample. Out of sample it collapsed. Here's why.